
import numpy as np              # Массивы (матрицы, векторы, линейная алгебра)
import matplotlib.pyplot as plt # Научная графика
%matplotlib inline 
    # Говорим jupyter'у, чтобы весь графический вывод был в браузере, а не в отдельном окне
import pandas as pd             # Таблицы и временные ряды (dataframe, series)
import seaborn as sns           # Еще больше красивой графики для визуализации данных
from sklearnex import patch_sklearn # оптимизация вычислений в sklearn (повышение быстродействия)
patch_sklearn()
import sklearn                  # Алгоритмы машинного обучения

Intel(R) Extension for Scikit-learn* enabled (https://github.com/intel/scikit-learn-intelex)


url = "australia_rain_tomorrow_raw.csv"
data_raw = pd.read_csv(url)


data_raw.shape

(142193, 24)


data_raw.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 142193 entries, 0 to 142192
Data columns (total 24 columns):
 #   Column         Non-Null Count   Dtype  
---  ------         --------------   -----  
 0   Date           142193 non-null  object 
 1   Location       142193 non-null  object 
 2   MinTemp        141556 non-null  float64
 3   MaxTemp        141871 non-null  float64
 4   Rainfall       140787 non-null  float64
 5   Evaporation    81350 non-null   float64
 6   Sunshine       74377 non-null   float64
 7   WindGustDir    132863 non-null  object 
 8   WindGustSpeed  132923 non-null  float64
 9   WindDir9am     132180 non-null  object 
 10  WindDir3pm     138415 non-null  object 
 11  WindSpeed9am   140845 non-null  float64
 12  WindSpeed3pm   139563 non-null  float64
 13  Humidity9am    140419 non-null  float64
 14  Humidity3pm    138583 non-null  float64
 15  Pressure9am    128179 non-null  float64
 16  Pressure3pm    128212 non-null  float64
 17  Cloud9am       88536 non-null   float64
 18  Cloud3pm       85099 non-null   float64
 19  Temp9am        141289 non-null  float64
 20  Temp3pm        139467 non-null  float64
 21  RainToday      140787 non-null  object 
 22  RISK_MM        142193 non-null  float64
 23  RainTomorrow   142193 non-null  object 
dtypes: float64(17), object(7)
memory usage: 26.0+ MB


data_raw.head()


# конвертирование типа столбцов в необходимые форматы
for col in data_raw.columns:
    col_type = data_raw[col].dtype
    n_unique = len(data_raw[col].unique()) - int(data_raw['RainToday'].isnull().values.any())
    if n_unique > 2 and col_type == 'O':
        data_raw[col] = data_raw[col].astype('category')
    elif n_unique == 2 and col_type == 'O':
        data_raw[col] = [1.0 if val == 'Yes' else 0.0 for val in data_raw[col]]
        data_raw[col] = data_raw[col].astype(np.float64)


data_raw.dtypes

Date             category
Location         category
MinTemp           float64
MaxTemp           float64
Rainfall          float64
Evaporation       float64
Sunshine          float64
WindGustDir      category
WindGustSpeed     float64
WindDir9am       category
WindDir3pm       category
WindSpeed9am      float64
WindSpeed3pm      float64
Humidity9am       float64
Humidity3pm       float64
Pressure9am       float64
Pressure3pm       float64
Cloud9am          float64
Cloud3pm          float64
Temp9am           float64
Temp3pm           float64
RainToday         float64
RISK_MM           float64
RainTomorrow      float64
dtype: object


data_raw.describe()


data_raw.describe(include=['category'])


# соотношение классов в бинарных признаках
print(data_raw.value_counts(subset=['RainToday'], normalize=True),
data_raw.value_counts(subset=['RainTomorrow'], normalize=True), sep = '\n\n')

RainToday
0.0          0.778787
1.0          0.221213
dtype: float64

RainTomorrow
0.0             0.775819
1.0             0.224181
dtype: float64


cols = data_raw.select_dtypes(include=[np.number]).columns
n_rows = int(np.sqrt(len(cols)) + 1)
n_cols = len(cols) // n_rows + 1
fig, axes = plt.subplots(nrows=n_rows, ncols=n_cols, figsize=(n_rows*4, n_cols*4))
for i, col in enumerate(cols):
    sns.histplot(data_raw, x=col, ax=axes[i // n_cols, i % n_cols], stat='density', hue='RainTomorrow')
    # фиксирование интервала для более понятного отображения
    if col == 'Rainfall':
        axes[i // n_cols, i % n_cols].set_ylim(0, 0.005)
        axes[i // n_cols, i % n_cols].set_xlim(0, 150)
    elif col == 'RISK_MM':
        axes[i // n_cols, i % n_cols].set_ylim(0, 0.025)
        axes[i // n_cols, i % n_cols].set_xlim(0, 150)
# удаление пустых графиков в последней строке
for i in range(len(cols), n_rows*n_cols):
    fig.delaxes(axes[i // n_cols, i % n_cols])


sns.pairplot(data_raw.sample(1000), hue='RainTomorrow')

<seaborn.axisgrid.PairGrid at 0x228417dc580>


fig, ax = plt.subplots(figsize=(12,12))
sns.heatmap(data_raw.corr(), vmin=-1, vmax=1, cmap='coolwarm', annot=True, fmt='.2f', ax=ax, square=True)

<AxesSubplot:>


for col in data_raw.columns:
    m_vals = data_raw[col].isnull().sum()
    p_m_vals = m_vals / data_raw.shape[0]
    print(col, m_vals, "{:.3f}".format(p_m_vals))

Date 0 0.000
Location 0 0.000
MinTemp 637 0.004
MaxTemp 322 0.002
Rainfall 1406 0.010
Evaporation 60843 0.428
Sunshine 67816 0.477
WindGustDir 9330 0.066
WindGustSpeed 9270 0.065
WindDir9am 10013 0.070
WindDir3pm 3778 0.027
WindSpeed9am 1348 0.009
WindSpeed3pm 2630 0.018
Humidity9am 1774 0.012
Humidity3pm 3610 0.025
Pressure9am 14014 0.099
Pressure3pm 13981 0.098
Cloud9am 53657 0.377
Cloud3pm 57094 0.402
Temp9am 904 0.006
Temp3pm 2726 0.019
RainToday 0 0.000
RISK_MM 0 0.000
RainTomorrow 0 0.000


data = data_raw.copy()
for col in data.columns:
    m_vals = data[col].isnull().sum()
    p_m_vals = m_vals / data_raw.shape[0]
    if p_m_vals <= 0.05:
        data = data.dropna(subset=[col], axis=0)


data.isnull().sum()

Date                 0
Location             0
MinTemp              0
MaxTemp              0
Rainfall             0
Evaporation      56101
Sunshine         61854
WindGustDir       6014
WindGustSpeed     6001
WindDir9am        7557
WindDir3pm           0
WindSpeed9am         0
WindSpeed3pm         0
Humidity9am          0
Humidity3pm          0
Pressure9am      11215
Pressure3pm      11151
Cloud9am         50169
Cloud3pm         51552
Temp9am              0
Temp3pm              0
RainToday            0
RISK_MM              0
RainTomorrow         0
dtype: int64


data.shape

(134590, 24)


import missingno as msno
msno.matrix(data)

<AxesSubplot:>


sorted_mv = data.sort_values('Sunshine')
msno.matrix(sorted_mv)

<AxesSubplot:>


sorted_mv = data.sort_values('Cloud9am')
msno.matrix(sorted_mv)

<AxesSubplot:>


msno.heatmap(data)

<AxesSubplot:>


msno.dendrogram(data)

<AxesSubplot:>


data.select_dtypes(include=['category']).isnull().sum()

Date              0
Location          0
WindGustDir    6014
WindDir9am     7557
WindDir3pm        0
dtype: int64


cat_cols = data.select_dtypes(include=['category']).columns
for col in cat_cols:
    data[col] = data[col].fillna(data[col].mode()[0])


data.select_dtypes(include=['category']).isnull().sum()

Date           0
Location       0
WindGustDir    0
WindDir9am     0
WindDir3pm     0
dtype: int64


cat_cols = data.select_dtypes(include=['category']).columns
cat_cols

Index(['Date', 'Location', 'WindGustDir', 'WindDir9am', 'WindDir3pm'], dtype='object')


from datetime import datetime


day_sin = []
day_cos = []
month_sin = []
month_cos = []
year = []
for date in data['Date']:
    f_date = datetime.strptime(date, '%d/%m/%Y')
    day_sin.append(np.sin(2 * np.pi * f_date.timetuple().tm_yday/365.0))
    day_cos.append(np.cos(2 * np.pi * f_date.timetuple().tm_yday/365.0))
    month_sin.append(np.sin(2 * np.pi * f_date.month/12.0))
    month_cos.append(np.cos(2 * np.pi * f_date.month/12.0))
    year.append(f_date.year)
new_date = pd.DataFrame({'day_sin': day_sin, 'day_cos': day_cos, 'month_sin': month_sin, 'month_cos': month_cos, 'year': year})
new_date


data.insert(0, 'year', year)
data.insert(0, 'month_cos', month_cos)
data.insert(0, 'month_sin', month_sin)
data.insert(0, 'day_cos', day_cos)
data.insert(0, 'day_sin', day_sin)
data.head()


data_fixed = data.copy(deep=True)
data_date = data['Date']
data['Date'] = pd.to_datetime(data['Date'])
data.set_index('Date', inplace=True)
data.sort_index(inplace=True)
data = data.reset_index()
data = data.drop('Date', axis=1)


data.head()


cat_cols = data.select_dtypes(include=['category']).columns


cat_cols.values

array(['Location', 'WindGustDir', 'WindDir9am', 'WindDir3pm'],
      dtype=object)


data.describe(include='category').loc['unique', :]

Location       49
WindGustDir    16
WindDir9am     16
WindDir3pm     16
Name: unique, dtype: object


from sklearn.preprocessing import OneHotEncoder

ohe_res = pd.DataFrame()
for col in cat_cols:
    ohe = OneHotEncoder()
    ohe_res = pd.concat([ohe_res, 
                         pd.DataFrame(
                             ohe.fit_transform(data[[col]]).toarray(), 
                             columns=[col + ' ' + str(i) for i in range(len(data[col].unique()))])], 
                        axis=1)
ohe_res


ohe_res.shape

(134590, 97)


data_transfromed = data.copy(deep=True)
data_transformed = pd.concat([data_transfromed.reset_index(), ohe_res.reset_index()], axis=1)


data_transformed.shape

(134590, 127)


data_transformed.head()


data_transformed = data_transformed.drop('index', axis=1)


data_transformed = data_transformed.drop(cat_cols, axis=1)


data_transformed.head()


full_data = data_transformed.copy(deep=True)
num_items = int(data_transformed.shape[0] / 100 * 20)
data_transformed = data_transformed.sample(num_items, random_state=42)


data_transformed.shape

(26918, 121)


from sklearn.impute import KNNImputer
imputer = KNNImputer(n_neighbors=3)
data_transformed = pd.DataFrame(imputer.fit_transform(data_transformed),columns = data_transformed.columns)


data_transformed.isnull().sum().unique()

array([0], dtype=int64)


data_copy = data.copy(deep=True)
data = data_transformed.copy(deep=True)
temp = data['RainTomorrow']
data = (data - data.mean(axis = 0))/data.std(axis = 0)
data['RainTomorrow'] = temp


y = data['RainTomorrow']
X = data.drop(['RainTomorrow'], axis=1)


# from sklearn.model_selection import train_test_split
# X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 42)
# 
# print(X_train.shape[0], X_test.shape[0])


from sklearn.model_selection import TimeSeriesSplit

tss = TimeSeriesSplit(n_splits = 4)
for train_index, test_index in tss.split(X):
    X_train, X_test = X.iloc[train_index, :], X.iloc[test_index,:]
    y_train, y_test = y.iloc[train_index], y.iloc[test_index]
print(X_train.shape[0], X_test.shape[0])

21535 5383


from tqdm.notebook import tqdm
from sklearn.neighbors import KNeighborsClassifier

train_score = []
test_score = []
for n_neighbors in tqdm(range(4, 26, 2)):
    model = KNeighborsClassifier(n_neighbors=n_neighbors)
    model.fit(X_train, y_train)
    train_score.append(model.score(X_train, y_train))
    test_score.append(model.score(X_test,y_test))


errors_dict = {
    'k_neighbors': {
        'n_neighbors': list(range(4, 26, 2)),
        'train_score': train_score,
        'test_score': test_score
    }
}
errors = pd.DataFrame(errors_dict['k_neighbors'], index=['k_neighbors, ' + str(i) for i in errors_dict['k_neighbors']['n_neighbors']])
errors = errors.drop('n_neighbors', axis=1)
errors


plt.plot(errors_dict['k_neighbors']['n_neighbors'], errors_dict['k_neighbors']['train_score'], color='green', marker='o', linestyle='solid')
plt.plot(errors_dict['k_neighbors']['n_neighbors'], errors_dict['k_neighbors']['test_score'], color='red', marker='o', linestyle='dashed')
plt.axhline(max(errors_dict['k_neighbors']['test_score']), color='purple')

<matplotlib.lines.Line2D at 0x2091496adc0>


max(errors_dict['k_neighbors']['test_score'])

0.8272338844510496


from sklearn.linear_model import LinearRegression

model = LinearRegression()
model.fit(X_train, y_train)

LinearRegression()


train_score = model.score(X_train, y_train)
test_score  = model.score(X_test, y_test)


errors_dict['Linear Regression'] = dict()
errors_dict['Linear Regression']['train_score'] = train_score
errors_dict['Linear Regression']['test_score'] = test_score
errors = pd.concat([errors, pd.DataFrame(errors_dict['Linear Regression'], index=['Linear Regression'])], axis=0)
errors


from sklearn.ensemble import RandomForestClassifier

model = RandomForestClassifier()
model.fit(X_train, y_train)

RandomForestClassifier()


pred = model.predict(X_test)
train_score = model.score(X_train, y_train)
test_score  = model.score(X_test, y_test)


errors_dict['Random Forest'] = dict()
errors_dict['Random Forest']['train_score'] = train_score
errors_dict['Random Forest']['test_score'] = test_score
errors = pd.concat([errors, pd.DataFrame(errors_dict['Random Forest'], index=['Random Forest'])], axis=0)
errors


data_fixed.head()


plt.plot(data_fixed['year'].sample(100, random_state=42).sort_index())

[<matplotlib.lines.Line2D at 0x26631f3c700>]


data_fixed['Date'] = pd.to_datetime(data_fixed['Date'])
data_fixed['Date'].head()

0   2008-01-12
1   2008-02-12
2   2008-03-12
3   2008-04-12
4   2008-05-12
Name: Date, dtype: datetime64[ns]


data_fixed.set_index('Date', inplace=True)
data_fixed.sort_index(inplace=True)
data_fixed.head()


data_fixed = data_fixed.reset_index()
data_fixed_date = data_fixed['Date']
data_fixed = data_fixed.drop('Date', axis=1)
data_fixed.head()


plt.plot(data_fixed['year'].sample(100, random_state=42).sort_index())

[<matplotlib.lines.Line2D at 0x266497841f0>]


data_fixed_Y = data_fixed['RainTomorrow']
data_fixed_X = data_fixed.drop('RainTomorrow', axis=1)


from sklearn.model_selection import train_test_split
data_fixed_train, data_fixed_test, _, _ = train_test_split(data_fixed_X, data_fixed_Y, test_size = 0.2, random_state = 42, shuffle=False)

print(data_fixed_train.shape, data_fixed_test.shape)

(107672, 27) (26918, 27)


fig, ax = plt.subplots(2, 1)
ax[0].plot(data_fixed_train['year'].sample(100, random_state=42).sort_index())
ax[1].plot(data_fixed_test['year'].sample(100, random_state=42).sort_index())

[<matplotlib.lines.Line2D at 0x26648f13040>]


from sklearn.model_selection import TimeSeriesSplit

tss = TimeSeriesSplit(n_splits = 3)
for train_index, test_index in tss.split(data_fixed_X):
    data_fixed_train, data_fixed_test = data_fixed_X.iloc[train_index, :], data_fixed_X.iloc[test_index,:]
    #y_train, y_test = y.iloc[train_index], y.iloc[test_index]
print(data_fixed_train.shape, data_fixed_test.shape)

(100943, 27) (33647, 27)


fig, ax = plt.subplots(2, 1)
ax[0].plot(data_fixed_train['year'].sample(100, random_state=42).sort_index())
ax[1].plot(data_fixed_test['year'].sample(100, random_state=42).sort_index())

[<matplotlib.lines.Line2D at 0x26648fa0af0>]


print(X_train.shape[0], X_test.shape[0], y_train.shape[0], y_test.shape[0])

21534 5384 21534 5384


from sklearn.neural_network import MLPClassifier


%%time
layers = [(100, 50), (50, 25), (25, 10), (10, 5), (100,), (50,), (25,), (10,)]
alphas = np.logspace(-3, 2, 15)
scores = {
    'n_layers': [],
    'n_neurons': [],
    'alpha': [],
    'train_score': [],
    'test_score': []
}
count = 0
for layer in tqdm(layers, desc='mlp progress'):
    for alpha in tqdm(alphas, desc='cycle progress'):
        mlp_model = MLPClassifier(hidden_layer_sizes = layer,
                                  alpha = alpha,
                                  solver = 'lbfgs', 
                                  activation = 'logistic',
                                  max_iter = 1000,
                                  random_state = 42)
        mlp_model.fit(X_train, y_train)
        scores['n_layers'].append(len(layer))
        scores['n_neurons'].append(layer)
        scores['alpha'].append(alpha)
        scores['train_score'].append(mlp_model.score(X_train, y_train))
        scores['test_score'].append(mlp_model.score(X_test, y_test))

Wall time: 14min 51s


mean_error = {
    'one-layered': {
        'mean_train_error_by_alpha': [],
        'mean_test_error_by_alpha': []
    },
    'two-layered': {
        'mean_train_error_by_alpha': [],
        'mean_test_error_by_alpha': []
    }
}
for i in range(len(alphas)):
    mean_error['one-layered']['mean_train_error_by_alpha'].append(np.mean(1 - np.array(scores['train_score'])[np.array(scores['n_layers']) == 1][i::len(alphas)]))
    mean_error['one-layered']['mean_test_error_by_alpha'].append(np.mean(1 - np.array(scores['test_score'])[np.array(scores['n_layers']) == 1][i::len(alphas)]))
    mean_error['two-layered']['mean_train_error_by_alpha'].append(np.mean(1 - np.array(scores['train_score'])[np.array(scores['n_layers']) > 1][i::len(alphas)]))
    mean_error['two-layered']['mean_test_error_by_alpha'].append(np.mean(1 - np.array(scores['test_score'])[np.array(scores['n_layers']) > 1][i::len(alphas)]))
    
fig, ax = plt.subplots(1, 2, figsize=(15, 8), sharey=True)
fig.suptitle('Mean. Error vs. alpha')
for i, layer in enumerate(['one-layered', 'two-layered']):
    ax[i].semilogx(alphas, mean_error[layer]['mean_train_error_by_alpha'], 'b-o', label = 'train')
    ax[i].semilogx(alphas, mean_error[layer]['mean_test_error_by_alpha'], 'r--o', label = 'test')
    ax[i].set_xlim([np.min(alphas), np.max(alphas)])
    ax[i].set_title(layer)
    ax[i].set_xlabel('alpha')
    ax[i].set_ylabel('error')
    ax[i].legend()


fig, ax = plt.subplots(1, 2, figsize=(15, 8), sharey=True)
fig.suptitle('Error vs. alpha')
for i, layer in enumerate(['one-layered', 'two-layered']):
    ax[i].semilogx(alphas, 1 - np.array(scores['train_score'])[np.array(scores['n_layers']) == (i+1)][:len(alphas)], 'b-o', label = 'train')
    ax[i].semilogx(alphas, 1 - np.array(scores['test_score'])[np.array(scores['n_layers']) == (i+1)][:len(alphas)], 'r--o', label = 'test')
    ax[i].set_xlim([np.min(alphas), np.max(alphas)])
    ax[i].set_title(layer)
    ax[i].set_xlabel('alpha')
    ax[i].set_ylabel('error')
    ax[i].legend()


tr_sc = np.array(scores['train_score'])
t_sc = np.array(scores['test_score'])
one_l = np.array(scores['n_layers']) == 1
two_l = np.array(scores['n_layers']) > 1
print('one-layered min error (train, test, optimal alpha): ',
      1 - np.max(tr_sc[one_l]),
      ', ',
      1 - np.max(t_sc[one_l]),
      ', ',
      alphas[np.where(t_sc[one_l] == np.max(t_sc[one_l]))[0][0] % len(alphas)],
      '\n',
      'two-layered min error (train, test, optimal alpha): ',
      1 - np.max(tr_sc[two_l]),
      ', ',
      1 - np.max(t_sc[two_l]),
      ', ',
      alphas[np.where(t_sc[two_l] == np.max(t_sc[two_l]))[0][0] % len(alphas)],
     sep='')

one-layered min error (train, test, optimal alpha): 0.0, 0.004086181277860312, 0.0610540229658533
two-layered min error (train, test, optimal alpha): 0.0, 0.0024145616641901846, 0.02682695795279726


mean_error = {
    'one-layered': {
        'mean_train_error_by_neurons': [],
        'mean_test_error_by_neurons': []
    },
    'two-layered': {
        'mean_train_error_by_neurons': [],
        'mean_test_error_by_neurons': []
    }
}
neurons = [np.sum(i) for i in scores['n_neurons'][::len(alphas)]]
for i in range(len(neurons) // 2):
    mean_error['one-layered']['mean_train_error_by_neurons'].append(np.mean(1 - np.array(scores['train_score'])[np.array(scores['n_layers']) == 1][i*len(alphas):(i+1)*len(alphas)]))
    mean_error['one-layered']['mean_test_error_by_neurons'].append(np.mean(1 - np.array(scores['test_score'])[np.array(scores['n_layers']) == 1][i*len(alphas):(i+1)*len(alphas)]))
    mean_error['two-layered']['mean_train_error_by_neurons'].append(np.mean(1 - np.array(scores['train_score'])[np.array(scores['n_layers']) > 1][i*len(alphas):(i+1)*len(alphas)]))
    mean_error['two-layered']['mean_test_error_by_neurons'].append(np.mean(1 - np.array(scores['test_score'])[np.array(scores['n_layers']) > 1][i*len(alphas):(i+1)*len(alphas)]))
    
fig, ax = plt.subplots(1, 2, figsize=(15, 8), sharey=True)
fig.suptitle('Mean. Error vs. neurons')
for i, layer in enumerate(['one-layered', 'two-layered']):
    ax[i].plot(neurons[i*len(neurons)//2:(i+1)*len(neurons)//2], mean_error[layer]['mean_train_error_by_neurons'], 'b-o', label = 'train')
    ax[i].plot(neurons[i*len(neurons)//2:(i+1)*len(neurons)//2], mean_error[layer]['mean_test_error_by_neurons'], 'r--o', label = 'test')
    ax[i].set_title(layer)
    ax[i].set_xlabel('neurons')
    ax[i].set_ylabel('error')
    ax[i].legend()


fig, ax = plt.subplots(1, 2, figsize=(15, 8), sharey=True)
fig.suptitle('Error vs. neurons')
nn_mask = np.full(np.array(scores['n_neurons']).shape, False)
nn_mask[::len(alphas)] = True
for i, layer in enumerate(['one-layered', 'two-layered']):
    ax[i].plot(neurons[i*len(neurons)//2:(i+1)*len(neurons)//2], 1 - np.array(scores['train_score'])[np.logical_and(nn_mask, np.array(scores['n_layers']) == (i+1))], 'b-o', label = 'train')
    ax[i].plot(neurons[i*len(neurons)//2:(i+1)*len(neurons)//2], 1 - np.array(scores['test_score'])[np.logical_and(nn_mask, np.array(scores['n_layers']) == (i+1))], 'r--o', label = 'test')
    ax[i].set_title(layer)
    ax[i].set_xlabel('neurons')
    ax[i].set_ylabel('error')
    ax[i].legend()

<ipython-input-336-bc54f5c44c4c>:3: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray
  nn_mask = np.full(np.array(scores['n_neurons']).shape, False)


tr_sc = np.array(scores['train_score'])
t_sc = np.array(scores['test_score'])
one_l = np.array(scores['n_layers']) == 1
two_l = np.array(scores['n_layers']) > 1
print('one-layered min error (train, test, optimal neurons): ',
      1 - np.max(tr_sc[one_l]),
      ', ',
      1 - np.max(t_sc[one_l]),
      ', ',
      neurons[np.where(t_sc[one_l] == np.max(t_sc[one_l]))[0][0] % len(neurons)],
      '\n',
      'two-layered min error (train, test, optimal neurons): ',
      1 - np.max(tr_sc[two_l]),
      ', ',
      1 - np.max(t_sc[two_l]),
      ', ',
      neurons[np.where(t_sc[two_l] == np.max(t_sc[two_l]))[0][0] % len(neurons)],
     sep='')

one-layered min error (train, test, optimal neurons): 0.0, 0.004086181277860312, 35
two-layered min error (train, test, optimal neurons): 0.0, 0.0024145616641901846, 75

	Date	Location	MinTemp	MaxTemp	Rainfall	Evaporation	Sunshine	WindGustDir	WindGustSpeed	WindDir9am	...	Humidity3pm	Pressure9am	Pressure3pm	Cloud9am	Cloud3pm	Temp9am	Temp3pm	RainToday	RISK_MM	RainTomorrow
0	01/12/2008	Albury	13.4	22.9	0.6	NaN	NaN	W	44.0	W	...	22.0	1007.7	1007.1	8.0	NaN	16.9	21.8	No	0.0	No
1	02/12/2008	Albury	7.4	25.1	0.0	NaN	NaN	WNW	44.0	NNW	...	25.0	1010.6	1007.8	NaN	NaN	17.2	24.3	No	0.0	No
2	03/12/2008	Albury	12.9	25.7	0.0	NaN	NaN	WSW	46.0	W	...	30.0	1007.6	1008.7	NaN	2.0	21.0	23.2	No	0.0	No
3	04/12/2008	Albury	9.2	28.0	0.0	NaN	NaN	NE	24.0	SE	...	16.0	1017.6	1012.8	NaN	NaN	18.1	26.5	No	1.0	No
4	05/12/2008	Albury	17.5	32.3	1.0	NaN	NaN	W	41.0	ENE	...	33.0	1010.8	1006.0	7.0	8.0	17.8	29.7	No	0.2	No

	MinTemp	MaxTemp	Rainfall	Evaporation	Sunshine	WindGustSpeed	WindSpeed9am	WindSpeed3pm	Humidity9am	Humidity3pm	Pressure9am	Pressure3pm	Cloud9am	Cloud3pm	Temp9am	Temp3pm	RainToday	RISK_MM	RainTomorrow
count	141556.000000	141871.000000	140787.000000	81350.000000	74377.000000	132923.000000	140845.000000	139563.000000	140419.000000	138583.000000	128179.000000	128212.000000	88536.000000	85099.000000	141289.000000	139467.000000	142193.000000	142193.000000	142193.000000
mean	12.186400	23.226784	2.349974	5.469824	7.624853	39.984292	14.001988	18.637576	68.843810	51.482606	1017.653758	1015.258204	4.437189	4.503167	16.987509	21.687235	0.221213	2.360682	0.224181
std	6.403283	7.117618	8.465173	4.188537	3.781525	13.588801	8.893337	8.803345	19.051293	20.797772	7.105476	7.036677	2.887016	2.720633	6.492838	6.937594	0.415065	8.477969	0.417043
min	-8.500000	-4.800000	0.000000	0.000000	0.000000	6.000000	0.000000	0.000000	0.000000	0.000000	980.500000	977.100000	0.000000	0.000000	-7.200000	-5.400000	0.000000	0.000000	0.000000
25%	7.600000	17.900000	0.000000	2.600000	4.900000	31.000000	7.000000	13.000000	57.000000	37.000000	1012.900000	1010.400000	1.000000	2.000000	12.300000	16.600000	0.000000	0.000000	0.000000
50%	12.000000	22.600000	0.000000	4.800000	8.500000	39.000000	13.000000	19.000000	70.000000	52.000000	1017.600000	1015.200000	5.000000	5.000000	16.700000	21.100000	0.000000	0.000000	0.000000
75%	16.800000	28.200000	0.800000	7.400000	10.600000	48.000000	19.000000	24.000000	83.000000	66.000000	1022.400000	1020.000000	7.000000	7.000000	21.600000	26.400000	0.000000	0.800000	0.000000
max	33.900000	48.100000	371.000000	145.000000	14.500000	135.000000	130.000000	87.000000	100.000000	100.000000	1041.000000	1039.600000	9.000000	9.000000	40.200000	46.700000	1.000000	371.000000	1.000000

	day_sin	day_cos	month_sin	month_cos	year
0	-0.478734	0.877960	-2.449294e-16	1.0	2008
1	-0.463550	0.886071	-2.449294e-16	1.0	2008
2	-0.448229	0.893919	-2.449294e-16	1.0	2008
3	-0.432776	0.901502	-2.449294e-16	1.0	2008
4	-0.417194	0.908818	-2.449294e-16	1.0	2008
...	...	...	...	...	...
134585	0.196673	-0.980469	1.224647e-16	-1.0	2017
134586	0.179767	-0.983709	1.224647e-16	-1.0	2017
134587	0.162807	-0.986658	1.224647e-16	-1.0	2017
134588	0.145799	-0.989314	1.224647e-16	-1.0	2017
134589	0.128748	-0.991677	1.224647e-16	-1.0	2017

	day_sin	day_cos	month_sin	month_cos	year	Date	Location	MinTemp	MaxTemp	Rainfall	...	Humidity3pm	Pressure9am	Pressure3pm	Cloud9am	Cloud3pm	Temp9am	Temp3pm	RISK_MM
0	-0.478734	0.877960	-2.449294e-16	1.0	2008	01/12/2008	Albury	13.4	22.9	0.6	...	22.0	1007.7	1007.1	8.0	NaN	16.9	21.8	0.0
1	-0.463550	0.886071	-2.449294e-16	1.0	2008	02/12/2008	Albury	7.4	25.1	0.0	...	25.0	1010.6	1007.8	NaN	NaN	17.2	24.3	0.0
2	-0.448229	0.893919	-2.449294e-16	1.0	2008	03/12/2008	Albury	12.9	25.7	0.0	...	30.0	1007.6	1008.7	NaN	2.0	21.0	23.2	0.0
3	-0.432776	0.901502	-2.449294e-16	1.0	2008	04/12/2008	Albury	9.2	28.0	0.0	...	16.0	1017.6	1012.8	NaN	NaN	18.1	26.5	1.0
4	-0.417194	0.908818	-2.449294e-16	1.0	2008	05/12/2008	Albury	17.5	32.3	1.0	...	33.0	1010.8	1006.0	7.0	8.0	17.8	29.7	0.2

	day_sin	day_cos	month_sin	month_cos	year	Location	MinTemp	MaxTemp	Rainfall	Evaporation	...	Humidity3pm	Pressure9am	Pressure3pm	Cloud9am	Cloud3pm	Temp9am	Temp3pm	RainToday	RISK_MM	RainTomorrow
0	-0.858764	0.512371	-5.000000e-01	0.866025	2007	Canberra	8.0	24.3	0.0	3.4	...	29.0	1019.7	1015.0	7.0	7.0	14.4	23.6	0.0	3.6	1.0
1	-0.493776	0.869589	-2.449294e-16	1.000000	2007	Canberra	15.1	20.4	22.6	2.4	...	90.0	1015.0	1014.1	8.0	8.0	17.0	16.3	1.0	4.2	1.0
2	-0.849817	0.527078	-5.000000e-01	0.866025	2007	Canberra	14.0	26.9	3.6	4.4	...	36.0	1012.4	1008.4	5.0	3.0	17.5	25.7	1.0	3.6	1.0
3	-0.478734	0.877960	-2.449294e-16	1.000000	2007	Canberra	11.6	26.3	4.2	1.6	...	46.0	1013.9	1009.5	2.0	2.0	16.7	25.2	1.0	0.2	0.0
4	-0.840618	0.541628	-5.000000e-01	0.866025	2007	Canberra	13.7	23.4	3.6	5.8	...	69.0	1009.5	1007.2	8.0	7.0	15.4	20.2	1.0	39.8	1.0

0. Задача¶

Homework 1¶

1. Чтение данных¶

2. Визуализация¶

Вывод¶

3. Пропущенные значения¶

4. Категориальные признаки¶

3. Пропущенные значения¶

5. Нормализация¶

6. Разбивка данных на выборки¶

7. Ближайшие соседи¶

8. Ошибки на выборках¶

Вывод¶

Вывод fixed¶

9. Другие классификаторы¶

Линейный классификатор¶

Случайный лес¶

Вывод¶

10. Несбалансированные классы¶

11. Корреляция¶

12. Вывод¶

Вывод fixed¶

Homework 1 FIX¶

Timeseries data wrong train_test split (shuffling old with young data)¶

Вывод¶

Homework 2¶

	Date	Location	WindGustDir	WindDir9am	WindDir3pm
count	142193	142193	132863	132180	138415
unique	3436	49	16	16	16
top	17/09/2014	Canberra	W	N	SE
freq	49	3418	9780	11393	10663

	train_score	test_score
k_neighbors, 4	0.851730	0.818874
k_neighbors, 6	0.843186	0.816831
k_neighbors, 8	0.839471	0.818317
k_neighbors, 10	0.834827	0.822404
k_neighbors, 12	0.830787	0.826491
k_neighbors, 14	0.828512	0.826491
k_neighbors, 16	0.827676	0.825562
k_neighbors, 18	0.827304	0.826491
k_neighbors, 20	0.824983	0.826491
k_neighbors, 22	0.824193	0.827234
k_neighbors, 24	0.822150	0.826677

	day_sin	day_cos	month_sin	month_cos	year	Location	MinTemp	MaxTemp	Rainfall	Evaporation	...	Humidity3pm	Pressure9am	Pressure3pm	Cloud9am	Cloud3pm	Temp9am	Temp3pm	RainToday	RISK_MM	RainTomorrow
Date
2007-01-11	-0.858764	0.512371	-5.000000e-01	0.866025	2007	Canberra	8.0	24.3	0.0	3.4	...	29.0	1019.7	1015.0	7.0	7.0	14.4	23.6	0.0	3.6	1.0
2007-01-12	-0.493776	0.869589	-2.449294e-16	1.000000	2007	Canberra	15.1	20.4	22.6	2.4	...	90.0	1015.0	1014.1	8.0	8.0	17.0	16.3	1.0	4.2	1.0
2007-02-11	-0.849817	0.527078	-5.000000e-01	0.866025	2007	Canberra	14.0	26.9	3.6	4.4	...	36.0	1012.4	1008.4	5.0	3.0	17.5	25.7	1.0	3.6	1.0
2007-02-12	-0.478734	0.877960	-2.449294e-16	1.000000	2007	Canberra	11.6	26.3	4.2	1.6	...	46.0	1013.9	1009.5	2.0	2.0	16.7	25.2	1.0	0.2	0.0
2007-03-11	-0.840618	0.541628	-5.000000e-01	0.866025	2007	Canberra	13.7	23.4	3.6	5.8	...	69.0	1009.5	1007.2	8.0	7.0	15.4	20.2	1.0	39.8	1.0